A crop gene function research scheme planning method based on experimental reasoning chain

Abstract

The application discloses a crop gene function research scheme planning method based on an experimental reasoning chain and belongs to the technical field of scientific research experiment intelligent planning. In view of the problems of lack of experimental reasoning chain modeling, insufficient search and recommendation capacity and low experimental scheme planning precision in the prior art, the application constructs a gene research knowledge graph by extracting research hypothesis-experimental operation-result interpretation triplets from historical documents, receives user experimental information, generates single-step experimental recommendations based on semantic vector matching and graph traversal, simultaneously generates a multi-round experimental scheme set through multi-scenario search, standardization processing and multi-dimensional score sorting, realizes the automatic planning of the crop gene function research experimental scheme, and improves the research efficiency and scientificity.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent planning of scientific research experiments, and particularly relates to a crop gene function research scheme planning method based on an experimental reasoning chain. BACKGROUND

[0002] Crop gene function research is an important task in the fields of modern agriculture and life science, and its core process usually follows the reasoning chain of research hypothesis, experimental operation and result interpretation. In actual research, scientists need to go through multiple rounds of experimental iteration. First, genetic material of the target gene is constructed or obtained and screened and preliminary phenotype analysis is performed. Then, based on the preliminary results, more complex experiments such as gene expression detection, subcellular localization, transcriptome or proteome analysis are designed to gradually reveal the molecular function and mechanism of the gene. However, this process has high complexity and uncertainty. Researchers need to refer to a large amount of previous research results, but existing academic literature is usually presented in a narrative form, lacking structured experimental reasoning chain records, which makes it difficult for researchers to efficiently utilize existing knowledge accumulation when formulating the next experimental scheme.

[0003] There are three main defects in the prior art. First, there is a lack of experimental reasoning chain modeling. Existing large language models and agents fail to systematically represent the research hypothesis, experimental operation and result interpretation process of crop gene function research, resulting in the inability to simulate and deduce at the research path level. Second, the retrieval and recommendation capability is insufficient. Existing technologies cannot efficiently retrieve and filter similar experimental cases in large-scale literature based on user-provided experimental results, and lack semantic comparison capability based on phenotypes or experimental phenomena. Third, the experimental scheme planning precision is low. The experimental steps recommended by existing models have a large gap with the real scientific research process, and lack context constraints and evidence-driven priority sorting mechanisms, which cannot effectively assist scientific research decision-making.

[0004] The existence of these defects is due to the lack of structured organization of scientific knowledge, and intelligent systems cannot understand and reproduce the reasoning logic of researchers. When faced with a large amount of unstructured literature, researchers need to spend a lot of time on manual sorting and comparison, and are likely to miss key information. At the same time, due to the lack of modeling of scientific logic chain, intelligent systems cannot accurately capture the logical association between experimental steps, resulting in recommended experimental schemes often lacking coherence and scientificity. These problems seriously restrict the efficiency and reliability of crop gene function research, and an intelligent experimental scheme planning technology that can systematically model the experimental reasoning chain is urgently needed. SUMMARY

[0005] To solve the above technical problems, the present application proposes a crop gene function research scheme planning method based on an experimental reasoning chain to solve the problems existing in the prior art.

[0006] In a first aspect, to achieve the above object, the present application provides a crop gene function research scheme planning method based on an experimental reasoning chain, comprising the following steps:

[0007] S1, refining the obtained historical literature into triples, and constructing a gene research knowledge graph based on the triples, wherein the triples include research hypotheses, experimental operations and result interpretations;

[0008] S2, receiving user input experimental information, traversing the gene research knowledge graph based on the experimental information, and generating a recommended next round of experimental scheme;

[0009] S3, based on the experimental information, traversing the gene research knowledge graph to obtain a plurality of similar situations, and after standardizing, comprehensively scoring and sorting the plurality of similar situations, generating a multi-round experimental scheme set.

[0010] Optionally, the process of S1 comprises:

[0011] Collecting literature data of crop gene function research from an academic database and preprocessing, converting the literature into a pure text format;

[0012] Using a large language model to automatically extract triples of research hypotheses, experimental operations and result interpretations from the preprocessed text;

[0013] Performing knowledge verification and quality control on the extracted triples, including logical consistency check and entity standardization;

[0014] Storing the verified triples as a knowledge graph, wherein the research hypotheses, experimental operations and result interpretations are nodes, and the logical relationships are edges.

[0015] Optionally, the process of S2 comprises:

[0016] Receiving user input of current experimental operation description and result interpretation text;

[0017] Using a semantic vector model to encode the result interpretation text into a query vector;

[0018] In the gene research knowledge graph, calculate the similarity of the query vector and all result interpretation nodes, and match the best similar node;

[0019] Based on the logical progressive relationship, traverse from the best similar node to the next research hypothesis node and the experimental operation node;

[0020] Output the recommended next research hypothesis and experimental operation as the next round of experimental scheme.

[0021] Optionally, the process of S3 comprises:

[0022] structurally analyzing the experiment information input by the user, extracting key information including core phenotype, gene and species entities;

[0023] based on the key information, retrieving a plurality of similar contexts in the gene research knowledge graph to obtain corresponding result interpretation nodes;

[0024] traversing from the result interpretation node of each similar context to the next experiment operation node to extract candidate experiment actions;

[0025] standardizing the candidate experiment actions, comprehensively scoring and ranking the standardized candidate experiment actions to generate a multi-round experiment scheme set.

[0026] Optionally, the standardization process includes:

[0027] identifying the original text expression from the extracted candidate experiment actions;

[0028] mapping the original text expression to the standard terms in the experiment method ontology;

[0029] replacing with standardized terms to form a unique candidate experiment action set.

[0030] Optionally, the comprehensive scoring and ranking process includes:

[0031] taking the frequency of each candidate experiment action appearing in the similar context as the literature evidence strength;

[0032] evaluating the matching degree of the candidate experiment action with the current research context of the user as the context constraint;

[0033] judging the consistency of the candidate experiment action with the scientific research promotion logic from phenomenon observation to mechanism analysis as the experimental target consistency;

[0034] based on the weighted sum of the literature evidence strength, the context constraint and the experimental target consistency to score;

[0035] ranking the candidate experiment actions in descending order of score to generate a ranked multi-round experiment scheme set.

[0036] In a second aspect, the present application also provides a crop gene function research scheme planning system based on an experiment reasoning chain, which is used to implement a crop gene function research scheme planning method based on an experiment reasoning chain, and the system includes:

[0037] a knowledge graph construction module, configured to extract historical literature into triples, and construct a gene research knowledge graph based on the triples, wherein the triples include research hypotheses, experiment operations and result interpretations;

[0038] The single-step recommendation module is configured to receive user inputted experiment information, perform traversal in the gene research knowledge graph based on the experiment information, and generate a recommended next round of experiment scheme;

[0039] The multi-round scheme planning module is configured to perform traversal in the gene research knowledge graph based on the experiment information to obtain a plurality of similar contexts, perform standardized processing, comprehensive scoring and sorting on the plurality of similar contexts, and generate a multi-round experiment scheme set.

[0040] In a third aspect, the present application further provides a computer terminal device, comprising:

[0041] one or more processors;

[0042] a memory coupled to the processor and configured to store one or more programs;

[0043] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the crop gene function research scheme planning method based on the experiment reasoning chain in the first aspect.

[0044] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the crop gene function research scheme planning method based on the experiment reasoning chain in the first aspect.

[0045] In a fifth aspect, the present application further provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the steps of the crop gene function research scheme planning method based on the experiment reasoning chain in the first aspect.

[0046] Compared with the prior art, the present application has the following advantages and technical effects:

[0047] The crop gene function research scheme planning method based on the experiment reasoning chain provided by the present application realizes systematic modeling and calculation of the crop gene function research process by constructing a structured experiment reasoning chain knowledge graph. The present application can automatically retrieve historical similar research contexts and generate a next step experiment scheme in accordance with scientific logic based on user inputted experiment information, thereby significantly improving the scientificity and efficiency of experiment planning. Through a multi-dimensional scoring and sorting mechanism, the present application ensures that the recommended scheme conforms to both the evidence strength of the literature and the specific research context, thereby enhancing the practicality and reliability of the recommended results. In addition, the present application supports cross-species experiment path recommendation, promotes effective utilization of research knowledge among different species, and provides intelligent decision support for crop gene function research. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application, and their

[0049] Figure 1 Flowchart of an embodiment of the application. DETAILED DESCRIPTION

[0050] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0051] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0052] Embodiment one

[0053] As shown in the accompanying drawings, the present embodiment provides a crop gene function research scheme planning method based on experimental reasoning chain, comprising: Figure 1

[0054] S1, refining the obtained historical literature into triples, and constructing a gene research knowledge graph based on the triples, wherein the triples include research hypothesis, experimental operation and result interpretation;

[0055] S2, receiving experimental information input by a user, traversing the gene research knowledge graph based on the experimental information, and generating a recommended next round of experimental scheme;

[0056] S3, based on the experimental information, traversing the gene research knowledge graph to obtain a plurality of similar situations, and after standardizing, comprehensively scoring and sorting the plurality of similar situations, generating a set of multi-round experimental schemes.

[0057] Specifically, it includes:

[0058] (1) Overview of system architecture and overall process:

[0059] Overall framework: based on the construction and utilization of PG-ERC knowledge graph, the present application assists in the automatic planning of crop gene function research scheme through the collaborative work of multiple intelligent agent modules. The system mainly consists of the following core modules:

[0060] PG-ERC knowledge graph;

[0061] SeedLLM, a large language model, and the HypothesisNavigator module;​

[0062] AdaptivePlanning module;

[0063] The core module and its operation process specifically include the following core technical solutions:

[0064] 1) PG-ERC knowledge graph construction:

[0065] The present application first constructs a knowledge graph (PG-ERC) for crop gene function research. The knowledge graph contains experimental reasoning chain data on crop gene function research from 1984 to 2024, including rice, corn, soybeans, and Arabidopsis thaliana. Each reasoning chain follows a three-part structure of "research hypothesis - experimental operation - result interpretation", covering about 10,000 genes and 120,000 experimental reasoning chains, supporting systematic retrieval and analysis of historical experimental paths.

[0066] The PG-ERC knowledge graph extracts the gene function research process scattered in massive literature into standardized "Hypothesis - Experiment - Conclusion" triples, and builds a knowledge graph based on these basic units. This structure completely preserves the logical chain of scientific discovery, not only revealing "what" (gene function), but more importantly, revealing "how to verify" and "why this conclusion is reached", providing deeper and more reliable reasoning basis for downstream applications (such as research assistance and breeding decision support).

[0067] 2) SeedLLM abundance large language model and HypothesisNavigator module:

[0068] HypothesisNavigator is an intelligent agent system based on SeedLLM abundance large language model. The fundamental purpose of its design is to activate and apply the PG-ERC knowledge graph constructed as described above. It is not a static knowledge base, but a dynamic and interactive research decision support engine, designed to automatically solve the critical iterative problem in gene function research: "What should be done next based on the current experimental results?".

[0069] The function of this module is to receive the user's completed experiments and their conclusions, and through intelligent interaction with the PG-ERC knowledge graph, automatically recommend, design and generate the optimal experimental scheme for the next round. It crystallizes the wisdom of millions of scientists in history (fixed in the logical chain of PG-ERC) into precise navigation of future research paths, directly serving the "evidence-driven iterative verification phase", achieving the intelligentization and dataization of research decisions.

[0070] 3) AdaptivePlanning module and multi-round experimental design:

[0071] Further, based on the inference of known experimental results, the AdaptivePlanning module uses historical experimental data to recommend experimental plans. This module analyzes the user's input information through structured analysis, performs semantic retrieval based on phenotypic similarity, extracts candidate "next experimental actions", and combines experimental method ontology for standardized processing. Finally, the system prioritizes and recommends candidate plans based on literature evidence strength, contextual constraints, and experimental goal consistency.

[0072] Through the above technical solutions, the present application not only realizes the construction of a gene function research inference chain from scratch and the automatic planning of a multi-round experimental path, but also provides cross-species experimental path recommendations between different species, significantly improving the efficiency and scientificity of plant gene function research.

[0073] As an embodiment in this embodiment, the process of S1 includes:

[0074] Collecting literature data on crop gene function research from academic databases and preprocessing, converting the literature to plain text format;

[0075] Using a large language model to automatically extract triples of research hypotheses, experimental operations and result interpretations from preprocessed text;

[0076] Knowledge verification and quality control of extracted triples, including logical consistency check and entity standardization;

[0077] Store the verified triples as a knowledge graph, where research hypotheses, experimental operations and result interpretations are nodes, and logical relationships are edges.

[0078] Specifically, it includes:

[0079] (2) Construction process of PG-ERC knowledge graph and scientific research logic chain:

[0080] 1) Construction process of PG-ERC knowledge graph:

[0081] The automatic construction process of PG-ERC is a systematic, phased data processing and knowledge engineering project, and its core is to convert unstructured scientific research literature into structured, computable knowledge. The entire process can be divided into the following five main stages:

[0082] First stage: Literature data collection and preprocessing:

[0083] This phase is the starting point of the entire process. First, systematically collect all scientific literature on rice, corn, soybeans, and Arabidopsis from 1984 to 2024 from major mainstream academic databases. These raw documents are converted into pure text format by the MinerU software parsing program. Then, the text is preprocessed, including cleaning irrelevant content (such as author information, references), dividing chapters, and cutting continuous text into independent sentences and words, preparing for the next stage of deep information extraction.

[0084] Second stage: Automatic extraction of ERC triple information:

[0085] This is the core link of the construction process, mainly executed by the specially trained SeedLLM large language model. In this stage, the preprocessed text is input into the SeedLLM. The model automatically scans and understands the scientific content of each sentence, accurately identifying key sentences or groups of sentences describing "research hypothesis (Hypothesis)", "experimental operation (Experiment)" and "result interpretation (Conclusion)". Through deep semantic analysis, the model will logically pair these three types of information to form an "ERC candidate set" containing a complete scientific reasoning chain. Each ERC unit corresponds to a specific gene function discovery process.

[0086] Third stage: Knowledge verification and quality control:

[0087] ERC candidate sets generated by automated extraction must undergo strict quality control processes to ensure the accuracy of the knowledge. This stage combines automated verification and manual review. Automated verification mainly uses different large language models to check the logical consistency of each ERC, and compares and standardizes the entities such as genes and species in it with authoritative biological databases. At the same time, the system will set a confidence threshold, mark ERCs below the threshold, and submit them to experts in the field of plant genetics for manual review and correction.

[0088] Fourth stage: Knowledge graph storage and construction:

[0089] High-quality ERC data screened through the quality control process will be officially used to build the knowledge graph. At this stage, each "hypothesis", "experiment", and "conclusion" is created as a "node" in the graph, and the logical relationship between them (such as "hypothesis" pointing to "experiment", "experiment" pointing to "conclusion") is created as an "edge". More importantly, the relationship between different ERCs (such as the "conclusion" of the previous experiment leading to the "new hypothesis" of the next study) is also established as an edge. Finally, these nodes and edges are stored in a professional graph database to form a large and precise PG-ERC network knowledge graph.

[0090] 2) PG-ERC scientific logic chain:

[0091] To accurately describe the scientific reasoning process extracted and constructed from a single document, a formal function expression is introduced. The complete scientific logic chain of an article ( ) can be represented by the following paradigm:

[0092] ;

[0093] Wherein, each symbol and constructor in the expression is defined as follows:

[0094] 3) : Basic Knowledge Nodes:

[0095] : represents the i-th research hypothesis (Hypothesis) node in the article. It encapsulates the research goal or scientific problem at this step.

[0096] : represents the i-th experimental operation (Experiment) node in the article. It encapsulates the specific experimental methods and processes performed to verify the hypothesis .

[0097] : represents the i-th result interpretation (Conclusion) node in the article. It encapsulates the direct conclusion drawn based on the experimental results.

[0098] The subscript i (taking values 1, 2,..., n) represents the logical order of ERC in the literature narrative.

[0099] 4) T( ): ERC Unit Constructor:

[0100] T is a function that takes a hypothesis node, an experiment node, and a conclusion node as input.

[0101] Its function is to construct a structured ERC (Experimental Reasoning Chain) basic unit.

[0102] Within this unit, two necessary internal directed connections are implicitly defined: one from... point to ( → ), representing "hypothesis-driven experiment"; another from point to ( → This represents "the experiment produces a conclusion." This constitutes a self-contained, minimal scientific research logical loop.

[0103] 5) »: Logical Progression Operator:

[0104] » is a core binary operator used to connect two adjacent ERC units.

[0105] Its operation rules are: A path will be created in the knowledge graph starting from the conclusion node of the previous unit. Hypothetical node pointing to the next cell External directed connections → .

[0106] Function: This operator formally defines the iteration and evolution of scientific research. It precisely represents that "the conclusion of the previous experiment is the direct basis for the next new research hypothesis," thus linking independent ERC units into a complete and coherent chain of scientific reasoning.

[0107] In this way, PG-ERC strings together the research history of a gene in one or more papers into a clear, traceable, and reasonable knowledge chain, which is its core advantage and value compared to traditional knowledge graphs.

[0108] As one implementation method in this embodiment, the process of S2 includes:

[0109] Receive user input of a description of the current experimental operation and an interpretation of the results;

[0110] The semantic vector model is used to encode the interpreted text of the results into query vectors;

[0111] In the gene research knowledge graph, the similarity of the calculation query vector and all result interpretation nodes is calculated, and the best similar node is matched;

[0112] Based on the logical progressive relationship, the next research hypothesis node and experimental operation node are traversed from the best similar node;

[0113] Output the recommended next research hypothesis and experimental operation as the next round of experimental scheme.

[0114] Specifically includes:

[0115] (3) The composition and automated workflow of SeedLLM and HypothesisNavigator modules:

[0116] The workflow of HypothesisNavigator is an automated process combining semantic vector retrieval and graph reasoning, and its core is to map the user's "current problem" to the "historical optimal solution" in the knowledge graph. The entire process can be divided into the following five main stages:

[0117] First stage: user input and query construction:

[0118] This stage is the starting point of human-computer interaction. The user inputs two core information through the system interface: the description of the current completed experimental operation E user and the key results and conclusions C user obtained from the experiment. The system integrates these two parts of text into a semantically complete "current research status" query.

[0119] Second stage: query semantic vectorization:

[0120] This is a key step to convert the user's natural language query into a machine-understandable vector. The system calls the advanced bge-m3 (BAAI General Embedding) vector model to encode the user's input text query into a high-dimensional query vector V q . This vector can accurately capture the deep scientific semantics of the user's current research progress, beyond the limitations of traditional keyword matching.

[0121] Third stage: knowledge retrieval and similarity matching:

[0122] The core of this stage is to efficiently find the most similar reference point in the vast amount of historical experience of PG-ERC to the user's current situation. During the construction of PG-ERC graph, the text content of all "conclusion" nodes has been pre-calculated and stored by the bge-m3 model. The system uses the user's query vector V qTo probe, perform efficient similarity search in the vector bank of tens of thousands of conclusion nodes, find the one that is semantically closest to the query The one that is semantically closest to the query can be called the best match conclusion .

[0123] Stage 4: Graph reasoning and path traversal

[0124] Once the best match conclusion is located , the system will use the most core graph structure of the PG-ERC knowledge graph to perform logical reasoning. According to the aforementioned definition of the scientific research logic chain , there must be a directed edge created by the logical progression operator (») in the system, pointing from C k to the next research hypothesis H k+1 . The reasoning module will automatically traverse this edge and accurately locate the "next round of research hypothesis" H k+1 and the "next round of experimental design" E k+1 proposed in the historical context.

[0125] Stage 5: Recommended scheme generation and output

[0126] The system will take the text content of H k+1 and E k+1 obtained by reasoning as core materials, integrate, polish, and format them through the SeedLLM large language model, and finally generate a user-friendly, logically clear, and specific next round of experimental recommendation scheme, and present it to the user.

[0127] In order to accurately describe the automatic recommendation process of the HypothesisNavigator module, its core function is defined as a query function Recommend_Next_Step. This function receives the user's current experimental progress, queries the PG-ERC knowledge graph, and returns the recommended next step.

[0128] Its function can be represented by the following paradigm:

[0129] ;

[0130] Wherein, the symbols and internal operations in this expression are defined as follows:

[0131] Input parameters (Input Parameters):

[0132] C user : User input text string describing the current experimental results and conclusions. To simplify the representation, the conclusion is taken as the core semantic anchor for the query.

[0133] G PG-ERC : Complete PG-ERC knowledge graph instance, containing all nodes and edges defined by » operator.

[0134] Output Results:

[0135] : System recommended next round of research hypothesis (Recommended Hypothesis) node content.

[0136] : System recommended next round of experimental operation (Recommended Experiment) node content.

[0137] Internal Operations:

[0138] Vectorization: V q = Vectorize_bge_m3(C user );

[0139] This step uses the bge-m3 model to convert the user input conclusion text C user into a query vector V q .

[0140] Similarity Search:

[0141] C k = ∈ GPG-ERC CosineSimilarity(V q , Vector( ));

[0142] This step finds the node with the highest cosine similarity with the query vector V q among the pre-stored vectors Vector( ) of all conclusion nodes in G PG-ERC , and records it as .

[0143] Graph Traversal:

[0144] H rec = Find_Successor(C k , via="»");

[0145] E rec = Find_Successor(H rec , via="→");

[0146] This step first defines the C→H external connection using the » operator, from C k to find its successor hypothesis node H rec . Then, it defines the H→E internal connection using the T constructor, from H rec to find its successor experiment node E rec .

[0147] Through these formal definitions, the HypothesisNavigator transforms the expert-level scientific decision-making process into a rigorous, reproducible, and efficient computational process, serving as a key bridge from "knowledge storage" to "intelligent application".

[0148] As an embodiment in the embodiment, the process of S3 comprises:

[0149] Structurally analyzing the experiment information input by the user, extracting key information, including core phenotypes, genes, and species entities;

[0150] Based on the key information, retrieving a number of similar scenarios in the gene research knowledge graph to obtain corresponding result interpretation nodes;

[0151] Traversing from the result interpretation node of each similar scenario to the next experiment operation node to extract candidate experiment actions;

[0152] Standardizing the candidate experiment actions, comprehensively scoring and ranking the standardized candidate experiment actions, and generating a set of multi-round experiment schemes.

[0153] Specifically, it comprises:

[0154] (4) System composition and workflow of the AdaptivePlanning module:

[0155] The workflow of AdaptivePlanning is a multi-stage, multi-dimensional information processing and decision-making process, which upgrades the simple "search-recommendation" to a complete closed loop of "search-generate-evaluate-sort".

[0156] First stage: structured information analysis:

[0157] This stage is the basis for accurately understanding the user's research status. The user inputs not only the final paper text but also multi-dimensional information including phenotype description, experimental materials, gene ID, etc. The system first structures the input and uses named entity recognition (NER) and other technologies to accurately extract key information entities such as key phenotype, gene, and species.

[0158] Second stage: candidate context retrieval based on phenotype similarity:

[0159] This is the source of candidate scheme generation. The system uses the "core phenotype" description parsed in the first stage as the main query content and uses the bge-m3 model to vectorize it. Then, a semantic similarity search is performed in the vector library of all nodes in the PG-ERC knowledge graph. Unlike HypothesisNavigator, the goal of this stage is not to find a unique best match, but to retrieve a batch of historical research contexts (i.e., multiple different Conclusion nodes and their associated ERC units) that are highly similar in phenotype description.

[0160] As an embodiment in this embodiment, the process of standardization processing includes:

[0161] Identify the original text expression from the extracted candidate experimental actions;

[0162] Map the original text expression to standard terms in the experimental method ontology;

[0163] Replace with standardized terms to form a unique set of candidate experimental actions.

[0164] The specific process of standardization processing includes:

[0165] Third stage: candidate experimental action extraction and standardization:

[0166] The purpose of this stage is to induce all possible "next steps" from the retrieved multiple historical contexts. The system will traverse each similar context's conclusion node C k , and along the path of C k → H k+1 → E k+1 , extract its corresponding "next experimental action" E k+1 . This will result in a list of original experimental actions with repeated and diverse expressions. Then, standardization processing is performed on these original texts, such as mapping different expressions such as "real-time PCR", "qRT-PCR" to the standard term "RT-qPCR".

[0167] As an embodiment in this example, the process of comprehensive scoring and ranking includes:

[0168] The frequency of each candidate experimental action in similar situations as the strength of the literature evidence;

[0169] The matching degree of the candidate experimental action with the user's current research context as the context constraint;

[0170] The consistency of the candidate experimental action with the scientific research promotion logic from phenomenon observation to mechanism analysis as the experimental goal consistency;

[0171] Based on the weighted sum of the strength of the literature evidence, the context constraint and the experimental goal consistency, the score is calculated;

[0172] Rank the candidate experimental actions in descending order of score to generate a ranked multi-round experimental scheme set.

[0173] The specific process of comprehensive scoring and ranking includes:

[0174] Fourth stage: multi-dimensional priority ranking:

[0175] This is the core decision-making link of the AdaptivePlanning module. The system will evaluate and rank each standardized candidate experimental action through a comprehensive scoring model. The model mainly considers the following three dimensions:

[0176] (1) Evidence Strength: The frequency of the experimental action in similar historical situations. The higher the frequency, the more "mainstream" or "consensus" the choice is.

[0177] (2) Contextual Constraint: The matching degree of the experimental action with the user's current research context. For example, if the user is researching rice, the technology commonly used in rice is preferred; if the user has performed transcriptome analysis, the subsequent protein interaction experiment is logically more smooth.

[0178] (3) Goal Consistency: Whether the experimental action is consistent with the macro scientific research promotion logic from "phenomenon observation" to "mechanism analysis". For example, after preliminary phenotype confirmation, the goal consistency of gene expression analysis is usually higher than that of direct protein structure analysis.

[0179] Fifth stage: dynamic scheme set generation and output:

[0180] The system ranks all candidate experimental actions in descending order according to the integrated score of the fourth stage. Finally, a multi-round experimental plan set containing multiple (e.g., 4) alternatives is output. Each plan clearly lists the recommended experimental operation steps, the literature evidence sources (linked to specific paths in the PG-ERC) on which the recommendation is based, cross-species control information for reference, and other information to provide users with efficient, dynamic, and personalized decision support.

[0181] The core function of AdaptivePlanning can be formalized as a ranking function Rank_Adaptive_Plans, which receives structured user input and outputs a ranked list of experimental plans.

[0182] Ranked_Plan_Set = Rank_Adaptive_Plans(User_Input, G PG-ERC , N);

[0183] Input parameters:

[0184] User_Input: A structured object containing the user's current research information, such as {'phenotype': '...', 'gene': '...','species': '...'}.

[0185] G PG-ERC : The complete PG-ERC knowledge graph.

[0186] N: The number of alternative plans to be recommended (e.g., N = 4).

[0187] Output results:

[0188] Ranked_Plan_Set: A list of length N containing the recommended experimental plans ranked by priority.

[0189] Internal operation logic:

[0190] (1) Retrieve:

[0191] Context_Set = PhenotypeSearch(User_Input.phenotype, G PG-ERC );

[0192] Based on the phenotype description, return a set of similar historical conclusion nodes C k .

[0193] (2) Extract & Standardize:

[0194] Actions_raw = { Traverse(C k ) for C k in Context_Set};

[0195] Actions_std = Standardize(Actions_raw, Ontology_exp);

[0196] Extract next-step experiments E k+1 from each context, and standardize them using the ontology library, resulting in a set of unique candidate actions A.

[0197] (3) Scoring (Score):

[0198] Score(A) = Strength(A) + ContextFit(A, User_Input) + GoalAlign(A, User_Input);

[0199] For each candidate action A, a weighted sum is performed with a weight coefficient ( ).

[0200] (4) The scoring function calculates its priority.

[0201] Strength(A): Calculate the frequency of action A appearing in Actions_raw.

[0202] ContextFit(A, User_Input): Based on rules or models, judge the matching degree of A and user context (return a score between 0 and 1).

[0203] GoalAlign(A, User_Input): Based on the scientific research workflow model, judge the logical smoothness of A in the current stage (return a score between 0 and 1).

[0204] Ranking and selection (Rank & Select):

[0205] Ranked_Actions = Sort(Actions_std, key=Score, reverse=True);

[0206] Ranked_Plan_Set = Generate_Plans(Ranked_Actions[:N]);

[0207] All candidate actions are ranked in descending order according to the score, the top N are selected, and a complete plan is packaged for output.

[0208] This module performs deep structural analysis on user input information, performs more extensive semantic retrieval based on phenotype similarity in the PG-ERC knowledge graph, and extracts, generalizes and standardizes a series of candidate "next step experimental actions" from multiple similar historical scenarios.

[0209] Finally, the system prioritizes all candidate plans through a multi-dimensional evaluation model, providing a dynamic, personalized, multi-round experimental plan set containing multiple alternatives for the user.

[0210] In summary, the main contents provided by the present application include:

[0211] 1) Construction of crop gene function research experimental reasoning chain based on PG-ERC knowledge graph:

[0212] The present application constructs a PG-ERC (Plant Gene-Experimental Reasoning) knowledge graph for crop gene function research, extracts experimental reasoning chains from massive literature using automated information extraction technology, and encodes them in a "research hypothesis-experimental operation-result interpretation" three-part structure, providing a structured knowledge base for subsequent experimental reasoning and plan planning. The construction and application of this knowledge graph are the basic innovative points of the present application.

[0213] 2) Integration and application of SeedLLM and HypothesisNavigator modules:

[0214] The present application innovatively combines SeedLLM and HypothesisNavigator modules, and based on the structured data of PG-ERC knowledge graph, it performs multi-round experimental reasoning and design on the basis of automated literature retrieval and experimental plan recommendation. The SeedLLM module is responsible for extracting key information from literature, while the HypothesisNavigator module designs the next experimental plan based on the experimental reasoning chain. The core innovation lies in its ability to automatically generate experimental paths that conform to scientific logic for different gene function research.

[0215] 3) Multi-round experimental reasoning and personalized experimental plan recommendation of AdaptivePlanning module:

[0216] The AdaptivePlanning module is one of the key technologies of the present application, which can perform multi-round experimental reasoning based on known experimental results, and generate personalized experimental schemes according to input information, literature evidence strength and context constraints. This module not only supports cross-species experimental path recommendation, but also optimizes the recommended order of experimental schemes for different research goals. The innovation of this module lies in the ability to dynamically feedback and optimize the recommended scheme, gradually improving the recommendation accuracy.

[0217] 4) Intelligent output of experimental path and multi-species control:

[0218] Based on the experimental path recommendation, the present application can generate dynamic and executable experimental schemes by combining cross-species control information such as species annotation, homology mapping and phenotype ontology. This output function is unique in the prior art, as it not only considers the experimental data of the current species, but also draws on the research results of other species, which is helpful for cross-species gene function research.

[0219] 5) Standardization and priority sorting of experimental steps combined with experimental method ontology:

[0220] When recommending the next experimental scheme, the present application uses experimental method ontology for standardization, and combines literature evidence strength and experimental goal consistency for priority sorting. This innovation ensures that the recommended experimental scheme not only conforms to scientific logic, but also has high practical operability, which can meet the needs of different research scenarios.

[0221] Based on this, the crop gene function research scheme planning method based on the experimental reasoning chain provided by the embodiment of the present application, the present application realizes the systematic modeling and calculation of the crop gene function research process by constructing a structured experimental reasoning chain knowledge graph. The present application can automatically retrieve similar historical research situations based on the user input experimental information and generate the next step experimental scheme in accordance with scientific logic, which significantly improves the scientificity and efficiency of experimental planning. Through a multi-dimensional scoring and sorting mechanism, the present application ensures that the recommended scheme conforms to the literature evidence strength and fits the specific research context, enhancing the practicality and reliability of the recommended results. In addition, the present application supports cross-species experimental path recommendation, which promotes the effective use of research knowledge among different species and provides intelligent decision support for crop gene function research.

[0222] Embodiment two

[0223] In this embodiment, a computer terminal device is provided, comprising:

[0224] one or more processors;

[0225] a memory coupled to the processor, for storing one or more programs;

[0226] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the crop gene function research scheme planning method based on the experimental inference chain.

[0227] In this embodiment, a computer readable storage medium is also provided, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the crop gene function research scheme planning method based on the experimental inference chain.

[0228] In this embodiment, an electronic device is also provided, and the electronic device includes a memory and a processor, and the memory stores a computer program, and the processor is configured to run the computer program to implement the steps of the crop gene function research scheme planning method based on the experimental inference chain.

[0229] In this embodiment, a computer program product is also provided, and the computer program product includes a computer program, and the computer program is executed by a processor to implement the steps of the crop gene function research scheme planning method based on the experimental inference chain.

[0230] The above program can be run in the processor, or can also be stored in the memory (or called computer readable medium), and the computer readable medium includes permanent and non-permanent, removable and non-removable media, and can be realized by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0231] These computer programs can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer implemented process, and the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows Figure 1 The steps of the functions specified in one or more flows or one or more blocks can be implemented by different modules. Figure 1 The steps of the functions specified in one or more flows or one or more blocks can be implemented by different modules.

[0232] The embodiment provides such a device or system. The system is called a crop gene function research scheme planning system based on an experimental reasoning chain, and includes:

[0233] A knowledge graph construction module is configured to extract historical literature into triples, and construct a gene research knowledge graph based on the triples, wherein the triples include research hypotheses, experimental operations, and result interpretations;

[0234] A single-step recommendation module is configured to receive user-input experimental information, traverse the gene research knowledge graph based on the experimental information, and generate a recommended next-round experimental scheme;

[0235] A multi-round scheme planning module is configured to traverse the gene research knowledge graph based on the experimental information to obtain a plurality of similar situations, and generate a multi-round experimental scheme set after standardizing, comprehensively scoring, and sorting the plurality of similar situations.

[0236] As an implementation manner of the embodiment, the knowledge graph construction module includes:

[0237] A data acquisition unit is configured to acquire literature data of crop gene function research from an academic database and perform preprocessing, and convert the literature into a pure text format;

[0238] A triple extraction unit is configured to automatically extract triples of research hypotheses, experimental operations, and result interpretations from the preprocessed text using a large language model;

[0239] A quality checking unit is configured to perform knowledge checking and quality control on the extracted triples, including logical consistency checking and entity standardization;

[0240] A graph storage unit is configured to store the checked triples as a knowledge graph, wherein the research hypotheses, the experimental operations, and the result interpretations are nodes, and the logical relationships are edges.

[0241] As an implementation manner of the embodiment, the single-step recommendation module includes:

[0242] An input receiving unit is configured to receive user-input current experimental operation description and result interpretation text;

[0243] A vector encoding unit is configured to encode the result interpretation text into a query vector using a semantic vector model;

[0244] A similarity matching unit is configured to calculate the similarity between the query vector and all result interpretation nodes in the gene research knowledge graph, and match the best similar node;

[0245] a graph traversal unit configured to traverse from the best similar node to a next hypothesis node and an experimental operation node based on a logical progression relationship;

[0246] a scheme output unit configured to output the recommended next hypothesis and experimental operation as a next round of experimental scheme.

[0247] As an implementation in the embodiment, the multi-round scheme planning module comprises:

[0248] a parsing unit configured to structurally parse the experimental information input by the user, and extract core phenotype, gene and species entities;

[0249] a context retrieval unit configured to retrieve a plurality of similar contexts in the gene research knowledge graph based on the core phenotype description, and obtain corresponding result interpretation nodes;

[0250] an action extraction unit configured to traverse from the result interpretation nodes of each similar context to a next experimental operation node, and extract candidate experimental actions;

[0251] a standardization unit configured to standardize the candidate experimental actions, and use experimental method ontology to unify the terms;

[0252] a sorting unit configured to comprehensively score and sort the standardized candidate experimental actions, and generate a multi-round experimental scheme set.

[0253] As an implementation in the embodiment, the standardization unit comprises:

[0254] a text recognition sub-unit configured to recognize original text expressions from the extracted candidate experimental actions;

[0255] a term mapping sub-unit configured to map the original text expressions to standard terms in the experimental method ontology;

[0256] a replacement sub-unit configured to uniformly replace the original text expressions with the standardized terms to form a unique candidate experimental action set.

[0257] As an implementation in the embodiment, the sorting unit comprises:

[0258] a frequency calculation sub-unit configured to calculate the frequency of each candidate experimental action in the similar contexts as the strength of the literature evidence;

[0259] a matching degree evaluation sub-unit configured to evaluate the matching degree of the candidate experimental actions with the current research context of the user as the context constraint;

[0260] a consistency judgment sub-unit configured to judge the consistency of the candidate experimental actions with the scientific research promotion logic from phenomenon observation to mechanism analysis as the experimental target consistency;

[0261] a scoring subunit configured to score based on a weighted sum of the strength of the literature evidence, the contextual constraints, and the consistency with the experimental goal;

[0262] a ranking subunit configured to rank the candidate experimental actions in descending order of the scores, and generate a ranked set of multi-round experimental schemes.

[0263] The system or device is used to realize the functions of the methods in the above embodiments, each module in the system or device corresponds to each step in the method, which has been described in the method and will not be repeated here.

[0264] Through the above embodiments, the problem of crop gene function research scheme planning based on experimental reasoning chain in the related art is solved, thereby being able to guarantee to solve the problems existing in the prior art.

[0265] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A crop gene function research protocol planning method based on an experimental inference chain, characterized by, The method comprises the following steps: S1, refining the obtained historical literature into triples, and constructing a gene research knowledge graph based on the triples, wherein the triples include research hypotheses, experimental operations, and result interpretations; S2, receiving user input experimental information, traversing the gene research knowledge graph based on the experimental information, and generating a recommended next round of experimental scheme, the process comprising: receiving user input current experimental operation description and result interpretation text, and encoding the result interpretation text into a query vector; matching the best similar node in the gene research knowledge graph, traversing the gene research knowledge graph based on the logical relationship and the best similar node, and generating a recommended next round of experimental scheme; S3, based on the experimental information, traversing the gene research knowledge graph to obtain a plurality of similar situations, and after standardizing, comprehensively scoring and sorting a plurality of the similar situations, generating a multi-round experimental scheme set, the process comprising: structurally analyzing user input experimental information, extracting key information, and the key information including core phenotypes, genes, and species entities; based on the key information, retrieving a plurality of similar situations in the gene research knowledge graph to obtain corresponding result interpretation nodes, obtaining candidate experimental actions based on the result interpretation nodes, and after standardizing, comprehensively scoring and sorting the candidate experimental actions, generating a multi-round experimental scheme set.

2. The method of claim 1, wherein, The process of S1 comprises: collecting literature data of crop gene function research from an academic database and preprocessing, converting the literature into pure text format; using a large language model to automatically extract triples of research hypotheses, experimental operations and result interpretations from the preprocessed text; knowledge verification and quality control are performed on the extracted triples, including logical consistency check and entity standardization; store the verified triples as a knowledge graph, wherein research hypotheses, experimental operations and result interpretations are nodes, and logical relationships are edges.

3. The method of claim 1, wherein, The process of S2 comprises: using a semantic vector model to encode the result interpretation text into a query vector; in the gene research knowledge graph, calculate the similarity of the query vector and all result interpretation nodes, and match the best similar node; based on the logical progressive relationship, traverse from the best similar node to the next research hypothesis node and the experimental operation node; output the recommended next research hypothesis and experimental operation as the next round of experimental scheme.

4. The method of claim 1, wherein, The process of S3 comprises: traverse from the result interpretation node of each similar situation to the next experimental operation node to extract candidate experimental actions; standardize the candidate experimental actions, comprehensively score and sort the standardized candidate experimental actions, and generate a multi-round experimental scheme set.

5. The method of claim 4, wherein, The process of standardization comprises: identifying the original text expression from the extracted candidate experimental actions; mapping the original text expression to the standard terms in the experimental method ontology; replace with standardized terms to form a unique set of candidate experimental actions.

6. The method of claim 4, wherein, The process of comprehensive scoring and sorting comprises: the frequency of each candidate experimental action appearing in the similar situation is taken as the literature evidence strength; evaluate the matching degree of the candidate experimental action with the user's current research context as the context constraint; judging the consistency of the candidate experimental action with the scientific research promotion logic from mechanism analysis of phenomenon observation as experimental target consistency; scoring based on the weighted sum of literature evidence strength, context constraints and experimental target consistency; ranking the candidate experimental actions in descending order of scores to generate a set of ranked multi-round experimental schemes.

7. A crop gene function research protocol planning system based on an experimental inference chain, characterized by, The system for implementing the method of any one of claims 1-6 comprises: a knowledge graph construction module configured to extract the obtained historical literature into triples, and construct a gene research knowledge graph based on the triples, wherein the triples include research hypotheses, experimental operations and result interpretations; a single-step recommendation module configured to receive user input experimental information, traverse the gene research knowledge graph based on the experimental information, and generate a recommended next-round experimental scheme; a multi-round scheme planning module configured to traverse the gene research knowledge graph based on the experimental information to obtain several similar situations, perform standardized processing, comprehensive scoring and ranking on the similar situations, and generate a set of multi-round experimental schemes.

8. A computer terminal device, characterized by comprise: one or more processors; a memory coupled to the processors and configured to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1-6. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-6.

Images